Fixes for the mcp2221 gpiochip get_* calls
cover Fixes for the mcp2221 gpiochip get_* calls
The mcp2221 HID driver exposes a gpiochip device.
While gpioset seemed to work fine, gpioget always returned 1 on all 4 GPIOs of the component (0x01 for input in the field "direction", according to the documentation).
This patch series addresses this issue by fixing the order of the fields in the driver's representation of the chip answer, and adding consistency to the way the callbacks prepare their command and the way the hid event handler gets these fields.
The set callbacks use a similar mechanism, but work for now because setting a direction/value only requires gpio-specific positioning in the command preparation, and not in the raw_event handler.
The core of this issue being a discrepancy in the way the command and the answer fetch their fields of interest, I would also like to ask if we should uniformize a bit the way this driver handles gpio, and how. I thought about several possible implementations for that:
Use mcp->gp_idx as the base offset of the gpio for set too, and modify the raw_event handler to fetch all relevant data based on that; or set the buffers in the mcp structure as unions of the various commands handled and use gp_idx simply as the gpio index to access relevant data directly from the structured representation everywhere; or go back to ye old defines to ensure portability...
Louis Morhet (2):
HID: mcp2221: fix report layout for gpio get
HID: mcp2221: fix get and get_direction for gpio
drivers/hid/hid-mcp2221.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/2 HID: mcp2221: fix report layout for gpio get
The documentation of the component (section 3.1.12 GET GPIO VALUES) describes the hid report structure with two fields per gpio:
its value, followed by its direction.
However, the driver describes it with a wrong order:
direction followed by value.
Fix the structure representing the report answered by the chip to the GET GPIO VALUES command.
Fixes commit 567b8e9fed8a ("HID: mcp2221: Fix GPIO output handling")
--- drivers/hid/hid-mcp2221.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hid/hid-mcp2221.c +1 -1
@@ -79,8 +79,8 @@ struct mcp_get_gpio {u8 cmd;u8 dummy;struct { -u8 direction;u8 value; +u8 direction; } gpio[MCP_NGPIO ]; } __packed;
Signature · 1 line
2.17.1
PATCH 2/2 HID: mcp2221: fix get and get_direction for gpio
In reply to Louis MorhetThe mcp2221_raw_event retrieves the value and direction of gpio on the same command, by setting the value on mcp->status and the direction on mcp->gpio_dir; and the offset at which they are read is based on mcp->gp_idx, set by the gpiochip callbacks.
However, the individual gpiochip calls set the index to look for directly on the field they want to track. This create a "double offset" in the final read in the response report.
Align the behaviour of mcp2221_raw_event and mcp_gpio_get/mcp_gpio_get_direction by putting gp_idx on those calls to the base offset of the gpio status struct.
--- drivers/hid/hid-mcp2221.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/hid/hid-mcp2221.c +2 -2
@@ -594,7 +594,7 @@ static int mcp_gpio_get(struct gpio_chip *gc, mcp->txbuf[0 ] =MCP2221_GPIO_GET ; - mcp->gp_idx = offsetof(struct mcp_get_gpio, gpio[offset].value); + mcp->gp_idx = offsetof(struct mcp_get_gpio, gpio[offset]);mutex_lock (&mcp->lock); ret =mcp_send_data_req_status (mcp, mcp->txbuf,1 );
@@ -675,7 +675,7 @@ static int mcp_gpio_get_direction(struct gpio_chip *gc, mcp->txbuf[0 ] =MCP2221_GPIO_GET ; - mcp->gp_idx = offsetof(struct mcp_get_gpio, gpio[offset].direction); + mcp->gp_idx = offsetof(struct mcp_get_gpio, gpio[offset]);mutex_lock (&mcp->lock); ret =mcp_send_data_req_status (mcp, mcp->txbuf,1 );
Signature · 1 line
2.17.1
Benjamin Tissoires
AppliedQuoted 13 lines · Louis Morhet
The mcp2221 HID driver exposes a gpiochip device.
While gpioset seemed to work fine, gpioget always returned 1 on all 4 GPIOs of the component (0x01for input in the field "direction", according to the documentation).This patch series addresses this issue by fixing the order of the fields in the driver's representation of the chip answer, and adding consistency to the way the callbacks prepare their command and the way the hid event handler gets these fields.
The set callbacks use a similar mechanism, but work for now because setting a direction/value only requires gpio-specific positioning in the command preparation, and not in theraw_eventhandler.
Quoted 1 line
[...]
Applied to https://git.kernel.org/pub/scm/linux/kernel/git/hid/hid.git (for-6.4/mcp2221), thanks!
[1/2] HID: mcp2221: fix report layout for gpio get
https://git.kernel.org/hid/hid/c/e36c31f8cac5 [2/2] HID: mcp2221: fix get and get_direction for gpio https://git.kernel.org/hid/hid/c/ca6961d8a851
Cheers,
Signature · 1 line
Benjamin Tissoires <[email protected]>
Quoted 5 lines · Louis Morhet
The mcp2221 HID driver exposes a gpiochip device.
While gpioset seemed to work fine, gpioget always returned 1 on all 4 GPIOs of the component (0x01for input in the field "direction", according to the documentation).
Quoted 7 lines
This patch series addresses this issue by fixing the order of the fields in the driver's representation of the chip answer, and adding consistency to the way the callbacks prepare their command and the way the hid event handler gets these fields.
The set callbacks use a similar mechanism, but work for now because setting a direction/value only requires gpio-specific positioning in the command preparation, and not in theraw_eventhandler.
As you should have seen in the automatic replied, I took that series in because it seems to fix a rather worrying bug.
Quoted 11 lines
The core of this issue being a discrepancy in the way the command and the answer fetch their fields of interest, I would also like to ask if we should uniformize a bit the way this driver handles gpio, and how. I thought about several possible implementations for that:
Usemcp->gp_idxas the base offset of the gpio for set too, and modify theraw_eventhandler to fetch all relevant data based on that; or set the buffers in the mcp structure as unions of the various commands handled and usegp_idxsimply as the gpio index to access relevant data directly from the structured representation everywhere; or go back to ye old defines to ensure portability...
Honestly, it's hard to make a choice here. You haven't got a replied from the other mcp2221 folks in almost 10 days so I am not sure you'll get feedback directly.
May I suggest that you work on any one of these idea, and then submit a series? Generally, having the code ready makes it way easier to decide if this is a good solution or not, while having 3 different vague suggestions makes it hard to see the implications of them.
Also, just a side note, this driver is very limited in term of scope, as it only touches one dedicated device. Which means that whatever solution feels the more elegant to you has a good chance of being accepted :)
Cheers,
Benjamin
Quoted 15 lines
Louis Morhet (2):
HID: mcp2221: fix report layout for gpio get
HID: mcp2221: fix get andget_directionfor gpiodrivers/hid/hid-mcp2221.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)Signature · 1 line
2.17.1
Quoted 2 lines · Benjamin Tissoires
As you should have seen in the automatic replied, I took that series in because it seems to fix a rather worrying bug.
Thanks!
Quoted 3 lines
Also, just a side note, this driver is very limited in term of scope, as it only touches one dedicated device. Which means that whatever solution feels the more elegant to you has a good chance of being accepted :)
Well, using a struct to describe the layout of a message seems more elegant; but if I'm not mistaken, taking offsets of fields in a packed struct is UB... so I was surprised to even see that in Linux, that's why I was wondering if I should pursue in that direction.
--
Louis